Prabhav Singh
@psingh522
Followers
45
Following
7K
Media
3
Statuses
44
CS Graduate @JHUCompSci | Research @jhuclsp
Baltimore, MD
Joined October 2024
Unpopular opinion: Finding a simple idea that actually works is way harder than publishing a fancy one that kinda works. You have to fight the urge to overcomplicate, give up many fancier ideas, fail and pivot again and again until you hit the first principle that truly holds.
34
58
791
my goal in life is to join Anthropic, delete all try/except clauses from Claude’s training data, and then quit.
58
48
2K
Hard work doesn’t stop at the research lab — it extends onto the pitch!! 🍁⚽️
1
1
53
Switching to Stan ( https://t.co/8nnQgtErvk) after using Python for Bayesian modeling and inference is like finally seeing the light at the end of the tunnel lmao. Shit is amazing - simple, readable and fast. I have been missing out.
0
0
0
ReLU is essentially an element wise MoE with shared router and expert weights: y= I(w^Tx >= 0^Tx) w^Tx + I(w^Tx < 0^Tx) 0^Tx, where the two experts are parameterized by the w and 0 (zero) vectors. And then you only need to train the expert weights and router just follows along.
13
19
342
I'm in Rotterdam for #Interspeech2025! Presenting my papers: 🗓️ Count Your Speakers! Multitask Learning for Multimodal Speaker Diarization (Port 3, Oral) 🗓️ Emojudge: LLM Based Post-Hoc Refinement for Multimodal Speech Emotion Recognition (Dock 15, Poster)
0
0
1
I love it when an ML algorithm uses a billion for loops instead of libraries that run C/C++ in the background
The fact that Python is the standard for machine learning is a serious indictment of the field’s engineering standards
0
0
0
biggest gift to humanity any frontier lab could do is add a ton of uv examples in their posttraining mix 😁 save the people from python dependency management hell!!
11
27
356
some nerd noticed that the bellcurve in the meme was improperly shaped and calculated the true distribution
9
33
503
The end of American dominance of science? (But don’t worry! Artificial Intelligence is a priority. Though actually the request for AI funding in CISE is also less, but only 17% less, not 60-70% less.)
The US government is aiming to get rid of a quarter million people involved in science research and education by 2026. I have no words.
15
16
119
“My whole team, including those who were eager to apply for more permanent positions in the U.S., have no more interest,” says @JohnsHopkins’ @MathiasUnberath. Those looking for jobs are applying in Europe, “including some of my superstars.”
nytimes.com
As the United States cuts budgets and restricts immigration, China and Europe are offering researchers money and stability.
0
7
13
The United States has had a tremendous advantage in science and technology because it has been the consensus gathering point: the best students worldwide want to study and work in the US because that is where the best students are studying and working. 1/
3
56
270
Yes, sure. The Chinese PhD student coming Tsinghua and Indian PhD student coming from IIT Bombay are taking remedial math classes at Harvard.
REPORTER: Why do you not want the best and brightest from around the world come to Harvard? TRUMP: I do. But a lot of the people need remedial rath -- math. Did you see that?
13
5
244
Chain-of-thought reasoning is not enough! Pixel Reasoner incentivizes pixel-space reasoning with curiosity-driven RL
1
14
42
"EmoJudge: LLM Based Post-Hoc Refinement for Multimodal Speech Emotion Recognition" A submission to the SER in naturalistic conditions challenge - We propose a multimodal architecture followed by prompting an LLM with acoustic features to refine the output labels!
0
0
1
"Count Your Speakers! Multitask Learning for Multimodal Speaker Diarization" We introduce a multitask speaker counting loss that results in better embeddings and informed AHC for improved multimodal diarization.
1
0
1
Two papers I worked on over the last two semesters at @jhuclsp have been accepted to #Interspeech2025! Thanks to @jvillalba1980 for advising me on both papers!
2
1
5